| | --- |
| | license: cc-by-4.0 |
| | task_categories: |
| | - any-to-any |
| | language: |
| | - en |
| | tags: |
| | - continual learning |
| | --- |
| | |
| | # ContinuaL-NExT Benchmark Card |
| |
|
| | ## Dataset details |
| |
|
| | This benchmark is built upon a collection of widely used and publicly available multimodal datasets for both understanding and generation tasks, including VQAv2, ImageNet, Flickr30k, OCR-VQA, RefCOCO, and HQEdit. |
| |
|
| | This benchmark is adopted to evaluate the **multimodal continual learning** ability of **unified generation and understanding MLLMs**. |
| |
|
| | Specific information please kindly refer to our code (https://github.com/JingyangQiao/MAGE) and paper (Coming Soon). |
| |
|
| | ## Acknowledgement |
| | Some datasets (VQAv2, OCR-VQA, RefCOCO and ImageNet) in this benchmark are modified versions of **[CoIN]** by [Chen et al.] ([https://huggingface.co/datasets/Zacks-Chen/CoIN]), available under CC-BY-4.0. Modifications include adaptation and integration with new data to form a new benchmark. Full attribution to the original authors is maintained. We thank for the authors have made the contributions to the open-source community. |
